I finally found the time to further investigate this issue.
I have redesigned my LUN's so that only 85% of my storage pool are used. Now that there is still space available, I could create a new LUN and do the tests again with RDM.
There is clearly a difference using RDM or just a VMDK hard disk. Using RDM has much better performance.
I again created a ticket with EMC to address this issues. They state that this is normal because of the overhead. As I find this hard to believe, I would like to hear some other opinions.
Wscholz,
I will definitely look at that IO analyzer appliance. Looks pretty handy. Thx!